EU AI Act Day: My personal thoughts

I have been invited to give comments by a few on what I think of the EU AI Act.  I thought I would share here in case of wider interest so here goes…..

It has been great to see the importance of responsible and ethical AI come to the fore.  Introducing levels of governance and control to protect people from the harmful side of technology is obviously important.  The EU AI Act comes into force on the 1st August (today) and widens its net over coming months.  A lot of the focus is applied on the controls on the technology vendors however we dont believe enough impetus on the application of the technology exists.  With any technology there can be applications for good or for harm, depending on the scenario or use case you are applying the technology too, AI is one of such, Quantum will be another before long that requires similar levels of control.  As a result, firstly I advocate for Responsible Technology not just Responsible AI and there is a potential miss in the EU AI Act as a result.

Whilst there is much good in its intent there is the potential for the EU AI Act to also stifle innovation and place controls for European organisations developing the core models and technology as a result of the honourable intent to control carbon of the models also.  A double edged sword exists as such here but inevitably, and as we have seen recently from Meta, the legislation may place the whole of Europe on a back foot in being able to compete with the US, China or other countries on developing our own models or more worryingly as with the case of Meta us being able to leverage and use some of the latest and cutting edge models that ultimately can deliver societal changing solutions and benefit across every country and citizen in Europe.  It will be interesting to see how regulation adapts if Meta's position becomes a common ground with other providers.

Ultimately, applying controls and enforcement can only be a good thing, however, the need for Global interoperability is paramount, lets hope that others find a means for at least consistency in the application of similar law in countries that are commonly engaging with Europe such as the when the US and UK announce their inevitable equivalents.  We should not, even must not stifle innovation, but we must control the risk of AI in the technology but ultimately in how it is applied.  Swift adaptation of the law may well be needed as a result of some of the challenges outlined here. 

I would love to see more guidance on how controls and best practices should be applied and even how certain aspects should be measured to enable deployment and controls to be built to adhere to the EU AI Act over the coming months.  Organisations should also consider the broader aspects of other legislation such as the Data Protection Act, Digital Services Act, and other conflicting sometimes not easily interpreted legalities as they look at their AI strategies and deployments such to avoid fines.  That said most are still fighting with the potential for reputational damage in poorly architected or trained generative AI models.  

If you are interested in understanding or discussing more about this please do not hesitate to reach out.